The traditional structural design method based on mechanical analysis software has some limitations, such as low efficiency and expert experience reliance. The efficient automatic structural optimization design can be achieved by using intelligent algorithms. However, due to the random search feature, the optimization result and convergence are highly dependent on the parameter settings of the algorithm whose reasonable values need to be determined by the trial-and-error procedure. It results in inefficient optimization and substantial computational cost. Therefore, this paper introduces the multi-population collaboration and information sharing mechanism to improve such problems and its applicability in the structural optimization design is studied. The finite element model of a steel frame is built by MSC.Marc and the equivalent horizontal load from earthquake obtained by base shear method is exerted on the structure. The automatic optimization process is established based on finite element software and the intelligent algorithm with the aim of mininizing the total material cost of the structure. Multiple structural constraints are considered including the inter-story drift ratio, the stress ratio, and the stability and width-thickness ratio of the component. Several strategies are used to improve the performance of the genetic algorithm, such as the fitness scaling, the direction-based crossover operator, the non-uniform mutation operator, the adaptive probability, the elite strategy, the duplicate substitution mechanism, and the constraint-based strategy. Then the multi-population mechanism is introduced to such an algorithm. The results of different algorithms are compared with each other, which shows that the multi-population genetic algorithm can improve the dependence of optimization results on algorithm parameters and the efficiency of structural optimization design.